Brief

Researchers have developed BodyReLux, a novel framework that uses a diffusion-based approach to relight full-body human performances in videos with temporal consistency. The system is trained on a unique dataset combining traditional static capture with a dynamic performance capture method. By leveraging pretrained text-to-video models and a new token-based lighting conditioning technique, BodyReLux achieves photorealistic and robust video relighting, enabling dynamic control over lighting sequences. AI